Flexibility versus Formulaicity: Comparing Phrasal Verb Use between Human-written and AI-generated Academic Essays

Abstract

With the rise of Generative AI technology, numerous studies have compared linguistic features of human-produced and AI-generated writings. However, little attention has been paid to the use of phrasal verbs (PVs), a difficult type of two-part verbs, in academic writing. This study chose the 1.7-million-word arts and humanities essays from the British Academic Written English corpus and generated an equivalent AI-written corpus using ChatGPT3.5. Extracting PVs with an innovative dependency-based method, the authors compared the frequency, diversity, and disciplinary distribution of all the identified PVs, and examined the polysemy, semantic transparency, and collocations of the two PVs with the highest frequency in both corpora. Findings show that human writers used PVs approximately five times more than ChatGPT3.5 and demonstrated twice the PV diversity of ChatGPT3.5. There were also significant sub-disciplinary differences in both corpora, with essays in archaeology and linguistics using significantly fewer PVs than classics and comparative American studies. Regarding PV-specific comparison, humans used PVs with more meanings, more varied transparency, and broader collocations than ChatGPT3.5. Overall, this study revealed that human writing is more flexible, personal, and spontaneous, while AI writing is more formulaic, rigid, and predictable, which provides important implications for education, applied linguistics, and computer science.

Keywords

Full text

How to cite